Chapter 1 Introduction

1. Some nancial disasters 2. VaR: denition 3. Beyond VaR

Chapter 1. Introduction

1. Some nancial disasters

1
1.1

Some nancial disasters

Metallgesellschaft, Germany (1993) Procter & Gamble, USA (1994) Codelco, Chili (1993) Showa Shell, Japan (1994) Kashima oil, Japan (1995) China Aviation Oil, China (2004)

1.2

Financial institutions

Savings and Loans (S&L), USA (80s) Bank of England, Great Britain (1992) Bank Negara, Malaysia (1993) Crdit Lyonnais, France (1994) e Barings PLC, Great Britain (1995) Daiwa, Japan (1995) LTCM, USA (1998) Financial crisis in Asia (90s) Mizuho Securities, Japan (2005)

Value at Risk and the normal distribution/ab Universit Paris-Dauphine e

Chapter 1. Introduction

1. Some nancial disasters

Amaranth Advisors, USA (2006) Socit Gnrale, France (2008) ee e e US and World Financial crisis (2008-2009)

1.3

Government funds

Orange County, USA (1994)

1.4

Financial regulation

Professional bodies (ISDA,. . . ) Authorities of regulated exchanges (CME, NYSE-Euronext,. . . ) National regulators (SEC, . . . ) Governments and legislators (Group of 30,. . . ) Central Banks Bank for International Settlements (Basel, Switzerland)

Value at Risk and the normal distribution/ab Universit Paris-Dauphine e

Chapter 1. Introduction

2. VaR: denition

2
2.1

VaR: denition
Financial risks Risk and financial instruments currency risk interest rate risk equity risk commodity risk Risk and accounting operating or business risk asset risk translation risk Risk and trading market or trading risk liquidity risk credit risk operational risk legal risk systemic risk

Value at Risk and the normal distribution/ab Universit Paris-Dauphine e

Chapter 1. Introduction

2. VaR: denition

Market risk Uncertainty due to: uctuation of prices (volatility) of nancial assets and liabilities (interest rates, exchange rates, . . . ), variations in the correlations of prices (correlation risk, basis risk,. . . ) and its impact on present or future cashows.

Absolute risk: P&L measured in dollars. Relative risk: P&L measured relative to a benchmark (tracking error ). Usually the trading risk is a short term risk. It is the current risk when selling or buying nancial assets or writing liabilities.

Value at Risk and the normal distribution/ab Universit Paris-Dauphine e

Chapter 1. Introduction

2. VaR: denition

2.2

Market risk management

Objective : control market risk A tool : Value at Risk, VaR VaR is a measure of market risk under normal market conditions: unique global measure of risk at group level obtained from aggregation of market risk of relevant elements. Used for: nancial risk reporting: regulators, auditors, rating agencies, nancial press resource allocation: capital allocation between activities or entities based on risk and return performance evaluation: risk-adjusted measures of performance. Some downside to VaR: VaR dont deal properly with large losses of low probability (rare events) VaR are often obtained from estimates of variances and covariances, which are as volatile as the losses themselves VaR computations breaks down in period of market chaos VaR is not a coherent measure of risk (dened later).

Value at Risk and the normal distribution/ab Universit Paris-Dauphine e

Chapter 1. Introduction

2. VaR: denition

2.3

VaR

Portfolio For a stochastic process Xt , the subscript t is a date at which Xt is known. Xt is stochastic at date s < t, known at date s t. Ft is the information available at t. It includes Xts, s 0. Portfolio: long and/or short positions on one or more nancial instruments (stocks, bonds, derivatives,. . . ). Market value of a portfolio at date t: amount to receive/pay to transfer all claims and obligations to a third party. Sign convention: sign +: long position (amount to receive) sign -: short position (amount to pay) except otherwise stated, all position referred to in the seminar will be long. The market value of a portfolio is the algebraic sum of all its elements: Xt = X1t + X2t + . . . + XN t . The gross return or net gain or P &L on a portfolio between t et t + h is the sum of the variation in value of its N elements plus net revenues they generate: P &L = X1t+h X1t + . . . + XN t+h XN t + D1t(h) + . . . + DN t (h).

Value at Risk and the normal distribution/ab Universit Paris-Dauphine e

Chapter 1. Introduction

2. VaR: denition

The net loss L on a portfolio between t et t + h is the opposite of the variation in value of its N elements minus net revenues they generate: L = X1t X1t+h + . . . + XN t XN t+h D1t (h) + . . . DN t(h). If net revenues Dit (h) are added, with a proper sign, to Xit+h : P &L = = L = = X1t+h X1t + . . . + XN t+h XN t Xt+h Xt X1t X1t+h + . . . + XN t XN t+h Xt Xt+h .

P &L and L dened above are stochastic at date t. P &L > 0 or L < 0 : the gross return is a gain P &L < 0 or L > 0 : the gross return is a loss

Example: Stock portfolio

Xit = ni pit where pit is stock is market price and ni the number of shares i in the portfolio. ni > 0 : long position ni < 0 : short position P &L = n1 (p1t+h /a1t (h) p1t ) + . . . + nN (pN t+h /aN t (h) pN t ) +n1 d1t (h) + . . . + nN dN t (h) dit (h) per share dividends paid in the time interval [t, t + h] ait (h) adjustment coecients to compensate for nancial operations that aect shareholders equity accounts or the number of issued shares in the time interval [t, t + h].

Value at Risk and the normal distribution/ab Universit Paris-Dauphine e

Chapter 1. Introduction

2. VaR: denition

VaR Let Xt be the market price of a portfolio, where Xt is a stochastic process dened on some probability space with a ltration Ft which describes the information at each date t. Except otherwise stated, when computing the gross return Xt+h Xt , Xt+h includes the net revenue Dt (h). The conditional distribution function of Xt+h Xt given Ft is supposed to be continuous . Value at risk at t, for horizon h, at the condence level : value of the loss during the time interval [t, t + h] with a probability of not being exceeded: P (Xt Xt+h VaRt (h, ) | Ft ) = . Notation: P (. | Ft ) VaRt (h, ) is written Pt (.) is written VaR.

VaR is the (1 )-quantile of the conditional probability distribution of the gross return Xt+h Xt . Value at risk with respect to the mean: dened by the relation P (E(Xt+h | Ft ) Xt+h VaRmt (h, ) | Ft ) = . Notation: VaRmt (h, ) is written VaRm.

VaR: maximum loss in an orderly liquidation of the portfolio between t and t + h with a given condence level. Usual values: = 0.95 or 0.99, h = 1 day, 1 week, 2 weeks, 1 month.

Value at Risk and the normal distribution/ab Universit Paris-Dauphine e

Chapter 1. Introduction

10

2. VaR: denition

Graphical representation of a quantile density 6

1
A A A U -

-VaR

value Xt+h Xt

Use of returns Let Xt be the market price of a portfolio and Dt(h) the revenue generated in the time interval [t, t + h]. Rate of return, or simple return, of portfolio at t for horizon h: Xt+h Xt + Dt(h) Rt(h) = Xt Continuous rate of return, or log-return, of portfolio at t for horizon h: Xt+h + Dt (h) Rct (h) = log Xt Rt(h) and Rct (h) are stochastic. If the revenue Dt (h) is added to Xt+h with the proper sign: Xt+h Xt Xt+h Rt(h) = , Rct (h) = log . Xt Xt Rate of return for position i in the portfolio is dened in the same way when Xit = 0.

Value at Risk and the normal distribution/ab Universit Paris-Dauphine e

Chapter 1. Introduction

11

2. VaR: denition

Xt+h = Xt (1 + Rt(h)), Xt+h = Xt eRct(h) . From properties of the log function, most of the time: Rct (h) Rt(h) when h is small.

If Xt > 0 represents a long position Rt (h) and Rct(h) represent the net gain for $1 invested between date t and t + h. If Xt < 0 represents a short position Rt(h) and Rct (h) are just the percentage change in value. If Xt = 0 rate of return is not dened. It is the case for derivatives like OTC forward and swap contracts. In the eld of nance, statistical studies and large data basis use rate of returns, models are often based on rate of returns. If q is the -quantile of the rate of return and Xt > 0: Pt (Xt+h Xt VaR) = Pt(Rt (h) VaR/Xt ) = 1 that is VaR = Xt q1 .

Example.

If Rt (h) ; N (0.01; 0.0081), Xt = 10, 000 and = 0.95: q1 = 0.1380 VaR = 1380.

Value at Risk and the normal distribution/ab Universit Paris-Dauphine e

Chapter 1. Introduction

12

2. VaR: denition

Source: The Economist February 21st 2004.

Source: The Economist January 1st 2005.

Value at Risk and the normal distribution/ab Universit Paris-Dauphine e

Chapter 1. Introduction

13

2. VaR: denition

2.4

Discontinuous distribution function

General definition of a quantile When the conditional distribution function of Xt+h Xt is not continuous, VaRt (h, ) is chosen so that the dening probability is as close as possible to 1 . Let X be a one-dimensional random variable with distribution function F: F (x) = P (X x). Dene the -quantile q of X by: q = inf {F (x) }, 0 < < 1.
xR I

VaR at condence level is then dened as: VaR = q1 . Proposition 1.1 Let X be a one-dimensional random variable with distribution function F . 1) F (x) is left-continuous on ] , +[. 2) q is right-continuous on ]0, 1[. 3) A discontinuity of F (x) induces an interval where q is constant. 4) An interval where F (x) is constant induces a discontinuity of q . 5) If F is continuous strictly increasing, q is the unique solution of the equation: F (x) = .

Value at Risk and the normal distribution/ab Universit Paris-Dauphine e

Chapter 1. Introduction

14

3. Beyond VaR: axiomatic approach to risk measure

3
3.1

Beyond VaR: axiomatic approach to risk measure

Let U be a set of scalar random variables dened on the probability space (, A, P ) and : U I U represents a family of possible stochastic R. P&L. is: subadditive if X U, Y U, X + Y U = (X + Y ) (X) + (Y ) positively homogeneous if > 0, X U, X U = (X) = (X) translation adjusted (or translation invariant) if a I X U, X + a U = (X + a) = (X) a R, monotonous if X U, Y U, X Y = (X) (Y ) law invariant if X U, Y U, X = Y = (X) = (Y ).
law

Value at Risk and the normal distribution/ab Universit Paris-Dauphine e

Chapter 1. Introduction

15

3. Beyond VaR: axiomatic approach to risk measure

is a coherent measure of risk if is: subadditive positively homogeneous translation invariant monotonous

3.2

VaR is not coherent

Proposition 1.2 1) VaR is positively homogeneous, translation invariant, monotonous. 2) VaR is not subadditive. 3) Var is law invariant. A counterexample: VaR is not subadditive For any X scalar random variable let (X) = q1 where q1 is the (1 )-quantile of X. If X is the P&L of a portfolio on a given interval of time: V aR = (X).

Value at Risk and the normal distribution/ab Universit Paris-Dauphine e

Chapter 1. Introduction

16

3. Beyond VaR: axiomatic approach to risk measure

Let Xi, i = 1, 2, be the P&L of two portfolios on a given interval of time, and Xi = X1 + X2 the P&L of the merged portfolio. Suppose Xi, i = 1, 2, are iid with a Pareto distribution: F (xi) = (2 xi)1 1 if xi < 1 else

Moments of Xi, i = 1, 2, dont exist and if x < 4: P (X1 + X2 x) = We have: (Xi) = (1 )1 2 (X1 ) + (X2 ) < (X1 + X2). The last property shows that VaR is not subadditive. 2 2 log (3 x) . + 4x (4 x)2

Value at Risk and the normal distribution/ab Universit Paris-Dauphine e

Chapter 1. Introduction

17

3. Beyond VaR: axiomatic approach to risk measure

3.3

Consistent measures of risk

Let U be a set of scalar random variables dened on the probability space (, A, P ) and : U I As before U represents a family of possible R. stochastic P&L. is: positive if X U = (X) 0 and (X) = 0 if and only if X is certain. translation invariant if a I X U, X + a U = (X + a) = (X) R, positively homogeneous of degree 1 if > 0, X U, X U = (X) = (X) is a consistent measure of risk if is: positive translation invariant positively homogeneous of degree 1.

Value at Risk and the normal distribution/ab Universit Paris-Dauphine e

Chapter 1. Introduction

18

3. Beyond VaR: axiomatic approach to risk measure

3.4

Some risk measures

Let X be the P&L of a portfolio on a given interval of time, F (x) and q the distriubtion function and -quantile of the probability distribution. When expected values exist, below are some risk measures: Variance 2 = E((X E(X))2). Positive centered moments = (E((X E(X))2)) 2 . VaR VaR = q1 (X). Worst conditional expectation WCE = inf{E(X | A), A A, P (A) > 1 }. Tail conditional expectation TCE = E(X | X q1 ). Conditional VaR CVaR = inf{ Expected shortfall ES = 1 F (q1 ) E(X1 {Xq1 } ) 1 I q1 . 1 1 1 E(X s) s, s I R}. 1

Value at Risk and the normal distribution/ab Universit Paris-Dauphine e

Chapter 1. Introduction

19

3. Beyond VaR: axiomatic approach to risk measure

We conclude the Chapter with The following Proposition. Proposition 1.3 1) WCE and ES are coherent risk measures. 2) is a consistent risk measure.

Value at Risk and the normal distribution/ab Universit Paris-Dauphine e